So IBM introduced a new addition to its venerable POWER line of high-end RISC CPUs earlier in the month, and I think they got something for everybody in there. Its based on the Power 2.06 instruction set and offers a lot more of what made the POWER6 line great…and a lot less of one thing that didn’t. Ha..Ha…Of course you know I mean power consumption. IBM really got the power consumption down from where it was on the POWER6…In some places more than 50% for the high density configurations. This is a big deal for datacenters that are starting to reach the boundary of what their current power and cooling setup can handle. So by upgrading to POWER7 or rethinking their upcoming hardware purchases in terms of POWER7, these organizations can extend the life and capacity of their existing datacenters. This is going to let them put off that new UPS purchase or new 3-phase power buildout for a while.
But back to POWER7. This chip is made using the current IBM 45 nm process and contains approximately 1.2 billion transistors. I guess we’ll have to wait till POWER8 for the 32 nm process. My first impression is, boy this thing is BIG…It has a whopping 27 different instruction formats ! and hundreds of instructions that can run all cool special purpose vector and signal processing units in every nook and cranny on the die. I hope these guys remember what the R in RISC stands for !
Another thing you get over POWER 6 is more cores while still keeping that high clock frequency. Although POWER6 was great, the low number of cores per chip has been a common complaint. This has been rectified in POWER7 with up to 8 cores per chip. Each core has 2 integer pipes and four floating point units, so the unit can be executing up to 12 instructions per core on any one clock cycle. It also has a vector unit…which Ill come to in a second..And this new thingy that appeared in POWER6…The decimal floating point unit. This adds another floating point unit to the data path, but instead of operating on IEEE 754 base-2 numbers it designed to operate on decimals…Gone are the problems of representing 0.1 in binary. Of course, to make it all work you’ve got to use the Decimal128 data type in the IBM xlC or gcc advanced toolchain C/C++ compilers. But this will be a big help in financial data analysis applications.
The vector unit will also be helpful for visualization and other modeling. It is a SIMD instruction similar to Intel SSE with some nice additions. The vector register file contains 32 registers of 128 bits in length. As with all vector RISC machines, all vector and scalar operations are register-to-register. One nice feature is a whole set of vector-vector and vector-scalar operations that can do a multiply, negation and addition in a single instruction. There is even separate rounding instructions with 5 different types of rounding for both scalars and vectors. Closely associated with the vector unit is the Signal Processing Engine which implements instructions that are designed to compute the linear difference equations associated with DSP while leveraging the deep pipeline and multiple execution units on the chip.
There are many applications for this new processor and the line of POWER7 machines that run it…The high number of cores and memory architecture make it ideal for large-scale multi-terabyte data warehouses. With the proper data partitioning and SAN, queries can be configured to run using tens or hundreds of child processes that can really divide and conquer large scale analytics projects. On the other hand, the vector and signal processing components make this an ideal machine for simulation and scientific computing. I wonder if these bad-boys will make it into the next generation of Xbox ? Hmmm…Thanks for reading…
Earlier in the week Microsoft announced its new mobile operating system – Windows 7 Mobile. Hopefully this will give Windows-powered devices the shot in the arm they need to compete with the new Android and iPhone devices we are seeing. The Windows Mobile operating system has taken a beating lately and is looking awfully behind the times when compared with the newer mobile OSes. Browser based apps and non-mission critical applications on these new OSes are poised to take a significant chunk out of Microsoft’s hide. However Windows still reigns supreme as the mobile OS for field force automation type applications due to the robust synchronization capabilities of SQL Server and SQL Server Compact Edition. If you have an “occasionally connected” application that requires a portable database on the device, I really believe that Windows is your only answer – for now. But there are plenty organizations that have occasionally connected mobile apps deployed, and they are beginning to ask about porting to other OSes. I hope this newest introduction from Redmond will give them good reason to stay on Windows.
Organizations of various sizes from large to small are all trying to reduce their IT costs and provide increased performance. This may require them to finally abandon legacy systems, or invest in new infrastructure to provide higher levels of availability or performance. The database layer is one area that is typically targeted in these cost reductions. It’s a good place to start because in many cases, the database layer can be upgraded, consolidated or completely changed with little or no impact on the higher level application tiers that sit above it. Also, in most organizations a given database server may support many different applications, so improvements there can have wide reaching effects.
Database vendors are touting their clustering technology as a potential solution to these problems. A clustered database is comprised of multiple physical servers accessing a single shared disk that contains the data. This allows multiple less-expensive servers to act as a single large virtual server and to distribute load across all the machines. This provides higher levels of redundancy than possible with single machines because all the hardware is duplicated. One piece of equipment can fail and other nodes in the cluster can assume its responsibility. However, before we all jump on the cluster bandwagon, there are a few important things to note. Most importantly is that clustering is not right for every application. It helps to know a little about your load and the applications that will access the database to determine if a clustered database is right for you.
Oracle Real Application Clusters (RAC) is a clustering technology offered by Oracle that runs on both their Standard and Enterprise edition products. Oracle Guru Don Burleson has recently published an article on RAC best practices that covers the best practices one should follow when implementing RAC. As Burleson correctly states, the Achilles heel of any clustered database is the cluster interconnect. This is a private network connection between the cluster nodes that is used for synchronization, heartbeat and cache coherency. In the clustered database, each node has its own physical memory that is used to cache data so that recently used data can be read from memory instead of the disk. In a clustered situation, these two caches appear as a single unified cache to the database processes. If a table was just read on node 1 and it is in the cache, and a request is then issued on node 2 for the same table, that data will be sent across the interconnect to node 2 instead of being read from the shared disk.
The importance and impact of this behavior cannot be overstated and it gives us an important clue as to which applications are better suited for clustering and which are not. OLTP type application where queries operate on a small number of records on heavily indexed tables can realize a tremendous benefit from clustering. Since only small amounts of data are read at any one time, the amount of data that has to move through the interconnect is very small, and over time the distributed caches will contain the optimum configuration of the data for the queries that were run.
On the opposite end of the spectrum we have data warehousing and decision support type applications where long-running queries operate on data sets that are large when compared with the unified cache size. This type of application can bring the cluster to its knees rather quickly. Data warehouse operations typically consist of long-running queries that return many rows and may require several intermediate joins and sorts to produce the output. When one of these queries runs on a given node it will essentially flush the cache of all other data and fill it up with data from this query. Then when a request is made on another node for the same data, it must all be pushed through the interconnect. This will quickly saturate the interconnect and once that happens bad things are sure to occur. For example, since heartbeat messages are also sent over the cluster interconnect, saturation could result in nodes being booted from the cluster.
This makes RAC ideally suited for ERP, MRP and other operational systems within the enterprise. While data warehousing and decision support would be better suited to a single “big-box” solution. In either event, these examples are just some of the considerations that you must account for when contemplating a RAC database.
Cetova’s VP and expert in management reporting design, John Lambrianakos, compares JD Edwards World’s reporting tools with those in EnterpriseOne in his latest white paper Closing the Gap between JDE World and E1 Reporting, the subject of Alex Woodie’s article in the Four hundred Stuff newsletter. Both the article and the white paper are excellent reads.
The much-anticipated new generation of IBM’s Power chip line has arrived! Yes, IBM has kicked it up a notch, a pretty big notch actually on two levels: first, the expected technology, an impressive new lineup of high-end micro processor-based servers designed for industries requiring high up-time and heavy workloads, and IBM delivers all this power with less energy consumption; and second, marketing via social media: IBM is taking shots at Oracle/Sun Sparc in Facebook and YouTube under an election campaign theme of “Servers for Truth”. Check out the video below.
Thanks to Larry Dignan at zdnet for his reporting: Go to zdNet
Have you got a burning question about deploying mobile applications? Like, what’s the best mobile OS for you, or what’s the best strategy for keeping your field reps connected, or how can you save on mobile app costs and lower risk? Then you’ll want to connect with Dr. Arbeeny and his presentation: “Pushing JD Edwards to the Edge of the Network” at the IBM/Oracle/GL Associates/Snap event at the IBM offices in NYC on February 25th. Dr. Arbeeny is the master of mobile applications design, development and integration with JD Edwards EnterpriseOne and World. He heads up GL Associates’ Strategic Business Technologies Practice and specializes in technical architecture design, JD Edwards application integration, data warehouse deployment, custom systems design and development, and real-time marketing. Submit your questions before the event and/or schedule a private meeting. There are still a few seats available. Call Bill Glasofer at (201) 451 9121, xt 110, or email Bill Glasofer, or go to GL Associates for details.
Questions about JD Edwards product plans? The person you need to hear from is John Schiff, who leads the business development function for JD Edwards at Oracle, and he will be presenting at the IBM/Oracle/GL Associates/Snap event on February 25th in NYC. John is always informative and interesting, and he enjoys answering questions and providing customers with information to help them with their important decisions. There are still some seats available. Call Bill Glasofer at (201) 451 9121, xt 110, or email Bill Glasofer, or go to GL Associates for details.
Welcome! This is the place where GL Associates’ staff voices comments, shares insight, offers opinions, and shares interesting links and news about technology that will help your business.
